We compared historical stream fish community sampling and remote sensing data from 1993–2011 in an agriculturally-dominated drainage basin to determine the effects of varying levels of riparian and landscape disturbances on fish communities. Stream fish guilds were analyzed with National Land Cover Database and aerial imagery products to quantify disturbance levels at a local and watershed scale. Insectivores and benthic insectivores were associated with intact riparian areas, whereas environmentally tolerant and omnivorous species were found in environments with degraded riparian areas. Riparian disturbances had the greatest effect near the stream (0–10 m), and the effects diminished with increasing distance from the stream. This spatial scale corresponds with those reported in the literature whereby allocthonous inputs and riparian filtering influenced stream fish communities and water quality. The landscape disturbance was not a strong predictor for any fish guild, indicating that local riparian conditions were more biologically meaningful. Considering the rapid rates of land conversion to production agriculture across the midwestern United States, ensuring adequate riparian buffering of streams may help protect specialist aquatic species and their habitat. Given the broad temporal and spatial nature of our data, these results suggest that targeted conservation of the 10-m scale of intact riparian vegetation directly adjacent to streams is a sound practice and should be considered the minimum conservation goal for watershed restoration in low-gradient landscapes historically dominated by agricultural land-use practices.
It is widely acknowledged that intact riparian vegetation plays a role in shaping fish and invertebrate communities (Pusey and Arthington, 2003) along with maintaining water quality (Schlosser and Karr, 1981). Disturbance of riparian buffers, or removal of the established vegetative community that provides soil and bank stabilization, has been shown to increase sediment and nutrient loading, thereby potentially altering the stream communities (Waters, 1995). Riparian areas are often disturbed and degraded by watershed-scale activities (Kauffman et al., 1997). As agriculture is the dominant land use in many developed basins, landscape effects related to agricultural disturbances can be expected in these drainages (Allan, 2004). The present 5.4% annual conversion rate of grassland to corn/soybean production is comparable to the rapid conversion rates in the 1920s and 1930s, during the mass mechanization of American agriculture (Wright and Wimberly, 2013). As conversion continues, it is imperative we understand the effects of local riparian and landscape disturbances on fish communities so best-practices can be recommended for their protection.
Riparian vegetation prevents stream degradation by reducing sedimentation potential and filtering nutrients along with other harmful substances dissolved in runoff or adsorb to eroded sediment. Riparian buffer areas reduce the amount of sediment transported from agricultural lands (Yuan et al., 2009). Buffers work by reducing the velocity of runoff as it moves through streamside vegetation, decreasing entrainment capacity. Sediment subsequently falls out of entrainment as a function of reduced velocity (Leopold et al., 1964). Riparian cover is usually classified as grassy and woody vegetation, and both provide sediment filtering function at similar efficiencies (Lyons et al., 2000). Phosphorus and nitrogen, often limiting in aquatic communities, are common nutrients in agricultural runoff, and are removed by both types of vegetation. Woody vegetation usually results in higher nitrogen removal, whereas grassy vegetation more efficiently removes phosphorus from runoff (Lyons et al., 2000). Additionally, both vegetative classes also filter organic wastes, pesticides, heavy metals, and hydrocarbons (Wenger, 1999).
Channel morphology and fish habitat are also influenced by riparian vegetation. Grassy vegetation is associated with narrower channels, as grasses trap, store, and grow on sediments removed from runoff. Grassy riparian areas are more likely to provide undercut banks, which provide ideal habitat for many fish species (Hunt, 1993). In comparison, wooded areas have more channel dimension, substrate, and water velocity variation (Lyons et al., 2000). Both provide cover for fish. Woody riparian vegetation contributes large woody debris and root wads, and grassy or shrubby banks provide overhanging vegetation. Additionally, allocthonous inputs of arthropods can mediate stream community trophic cascades (Nakano et al., 1999), with higher vegetative diversity resulting in higher and more diverse arthropod inputs (Saunders and Fausch, 2007). A mosaic of both vegetative buffer types, providing a suite of functions and habitats, is a desirable riparian management goal (Schlosser and Karr, 1981, Lyons et al., 2000).
Stream fish assemblages are commonly assessed relative to the landscape by organizing the fish community into functional guilds. The index of biotic integrity (IBI) and associated fish guilds were developed by Karr (1981) as an alternative to single-species monitoring of stream status, a method with limited utility for explaining patterns and processes in stream ecology and condition (Thurston, 1979; Gosz, 1980; Karr and Dudley, 1981). Guilds remain useful to organize fish community structure and analyze it relative to the surrounding environment.
Given the influence of riparian vegetation on a variety of stream physical and chemical characteristics, our objectives were to evaluate links between the extent of intact riparian cover, landscape disturbance in the local watershed, and percent composition of fish guilds representing (1) feeding preference (omnivore, insectivore, piscivore), (2) feeding mode (benthic insectivore) and (3) environmental tolerance (tolerant species). The guilds represent functional groups that display sensitivity to lack of habitat diversity (omnivore, insectivore, piscivore), sedimentation (benthic insectivore) and degraded water quality (environmentally tolerant species). Through this work, we hoped to better understand the influence of riparian cover on fish community composition in agriculturally-dominated watersheds.
Our study region was the western tributaries of the Red River of the North within eastern North Dakota, USA, a drainage area of 34,000 km2 (Figure 1; Renard et al., 1986). Stream gradient varies from 0.04 to 0.25 m km−1 (Renard et al., 1986), with high levels of suspended sediments, primarily clays and silts that are relics of the Agassiz glacial lake plain (Stoner et al., 1993; Lyons, 2008). The mean total suspended load for the Red River Basin was calculated at 42 mg l−1 (Stoner et al.,1993). Approximately 71.5% of the basin was cultivated row crop agriculture, with a further 9.5% of the area occupied by grazing, totaling 81% agricultural land-use in the basin (Strong, 2010).
Archival fish sampling data were obtained from the North Dakota Department of Health and the North Dakota Game and Fish Department. Electrofishing sampling events occurred from 1993–2011 across the Red River basin (n = 274). Despite stream species detectability concerns (Reynolds, 1996), electrofishing data presented the largest temporal and spatial coverage of the study area. Abundances were converted to percent species compositions and pooled by guild: environmentally tolerant, omnivore, insectivore, benthic insectivore, and piscivore (Table 1). The species composition of the guilds was developed through the combined work of Niemela et al. (1998), Barbour et al. (1999), and Pflieger (1997).
Samples were grouped into three temporal bins (1992, 2006, 2011) named after the most temporally appropriate National Land Cover Database (NLCD) products available for the sampled periods. Data collection years 1993–1998 were grouped in the “1992” bin, years 2005–2007 were grouped in the “2006” bin, and years 2010–2011 were grouped in the “2011” bin. The bins were used to assign temporally-appropriate land cover and riparian vegetation metrics to each sampling location. Sampling replication was inconsistent spatially and temporally, with most locational replicates (n = 93) occurring within individual temporal bins. Replicate samples at a given location within each temporal bin were averaged, yielding a sample pool (n = 181) with one set of values at each site in a temporal bin. Within this pool, a subset of sites (n = 11) had replicates in a separate bin. For each of these pairs, sampling return interval was at minimum 10 years, environmentally tolerant, omnivore, insectivore, benthic insectivore guild percent compositions were uncorrelated (α = 0.10), and mean differences of landscape metrics reflected a dynamic landscape that had changed over time. We therefore treated these samples as independent, discrete observations included in the larger pool.
Intact riparian cover was digitized within a series of buffers (0–10, 10–20, 20–30, 30–40 and 40–50 m) perpendicular to each 3-km stream reach using the 1997–1998 USGS digital orthophoto quarter-quadrangles (DOQQ), 2006 and 2010 National Agriculture Imagery Program (NAIP) rasters. The 3-km stream reach reflects the hypothesized scale where riparian processes influenced stream conditions (Barton et al., 1985). For each 10-m buffer, the percent area of intact riparian cover was calculated. This yielded the area of intact riparian cover within 10-m bands out from the bank to a maximum of 50-m. Intact riparian cover was defined as areas adjacent to the stream that possessed undisturbed vegetation as determined from remote sensing imagery that would stabilize the bank area against erosion and could potentially function as a sheet runoff filter. Rip-rapped areas or other artificially stabilized areas were also classified as intact. The opposite was non-intact riparian area, or any area adjacent to the stream with naturally bare ground, cultivated agricultural land, vegetation removal, or other disturbances. Non-intact riparian cover included any additional areas that were not intact riparian cover, regardless of land-cover type. All geoprocessing was conducted in ArcMap 10.2.
The 1992 (Vogelmann et al., 2001), 2006 (Fry et al., 2011), and 2011 (Jin et al., 2013) NLCD full classification scheme was reduced to avoid spurious correlations. The NLCD products were reclassified into Disturbed, Undisturbed, and Open Water classes (Table 2). After reclassification, a stratified random sample of 30 points per land-cover class was applied and assessed in the ERDAS Imagine 2013 Accuracy Assessment module. The reference images were those used for the riparian digitizing (1997–1998 USGS DOQQ, 2006 NAIP, 2010 NAIP). The land-cover class accuracy minimum was 80.56%, and reclassification overall accuracies exceeded 90% in each temporal bin. Results of a Kappa analysis, analogous to a Chi-square test of land-cover classification error, all exceeded 0.80, indicating high producer accuracy (Congalton and Green, 2008). The reclassifications were therefore satisfactory, and no remedial action was necessary. The land-cover analysis was conducted at the subwatershed scale using the HUC-12 USGS catalog unit. This minimized pseudo-replication, as a larger catalog or catchment unit would have contained multiple sample reaches.
Initial averaging of replicate observations within temporal bins and use of the HUC-12 landscape scale reduced clustering of data points, which prevented spatial autocorrelation. Formal testing of spatial autocorrelation was implemented using Spatial Analysis in Macroecology (SAM) (Rangel, 2010) using Moran's I and was implemented for each of the analysis guilds. This also further evaluated the sample independence assumption described previously. Correlogram inflection points and Moran's I values were satisfactory, with interpretation following Legendre and Legendre (2012). Spatial autocorrelation was assumed to be inconsequential.
All further statistical analyses were conducted in R (R Core Team, 2013). Initial analyses focused on identifying correlational structure of predictor variables. Collinearity of the ring buffers was explored with correlation tests. Although all of the buffers were correlated (α = 0.05), the 10-m and 50-m buffers were the most weakly correlated (Pearson's r = 0.531) and were retained for analysis. A pooled buffer from 0–50 m was also used to investigate the vegetation gradient between the near- and distant-scale riparian cover. The disturbed land-cover class was chosen for analysis and the highly collinear inverse undisturbed class was discarded.
Multiple regressions using an information theoretic multi-model inference approach were used to determine if the 10-m riparian buffer, 50-m riparian buffer, and disturbed land-cover class influenced guild percent composition. A second, separate analysis was run using the pooled 0–50 m buffer and disturbed land-cover class was conducted to examine ecological scaling while reducing collinearity concerns. Normality and tolerance values were satisfactory. Models were ranked by ΔAICc, and a ΔAICc ≥4 threshold was used to differentiate candidate model performances from the intercept (null) model performance (Burnham and Anderson, 1998). Coarse evaluation of individual predictor support was assessed by summing AICc weights (ΣW) of all candidate models that included the predictor (MacKenzie et al., 2006). The ΣW scale is 0–1, with 0 offering no support and 1 offering substantial support (MacKenzie et al., 2006). Model-averaged estimates from the 95% confidence model set were generated to understand the landscape predictor effects. Multi-model inference and model averaging used R library “AICcmodavg” (Mazerolle, 2014).
Candidate models for the environmentally tolerant, omnivore, insectivore, and benthic insectivore guilds displayed meaningful relationships with landscape predictors and consistently included the 10-m riparian term. Most models that included the 10-m term performed better than the intercept model for the given guild. Evidence suggested that the proportion of intact vegetation in the near-stream riparian scale was important across several guilds (Table 3). The 50-m riparian term was meaningful for the insectivorous guild, but overall the independent 50-m riparian term model did not perform as well as the 10-m riparian model. The proportion of disturbed watershed land-cover did not perform well independently, but was included in top-performing models for each guild.
Summed model weights for each term gave further evidence for the 10-m riparian term effect and helped dissect the importance of each predictor (Table 4). For the environmentally tolerant, omnivore, insectivore, and benthic insectivore guilds, the summed 10-m riparian model weights were notably larger than the 50-m, disturbed land-cover, and intercept weights, evincing a greater effect relative to the other predictors. The 50-m model weights were generally the result of inclusion in models containing the 10-m term. The insectivore guild showed evidence for an independent 50-m effect, but the sum of weights was half of the 10-m effect.
Piscivore percent composition was best described by the null candidate model, indicating the guild responded to neither riparian vegetation nor landscape disturbance in this dataset. It is unlikely that these landscape metrics have no effect on piscivores, but data paucity prevented any meaningful results. As result, no model averaged estimate was generated.
Model averaged estimates were calculated from the 95% confidence candidate model set. Model averaged estimates offered additional support for a 10-m riparian effect for the environmentally tolerant, omnivore, insectivore, and benthic insectivore guilds (Table 5). If the confidence interval for a result included zero, the estimate was considered to have no support. Based on this, the environmentally tolerant, omnivore, insectivore, and benthic insectivore guilds responded to neither 50-m riparian vegetation nor the landscape disturbance.
The tolerant and omnivorous guilds displayed negative responses, whereas the insectivorous and benthic insectivorous guilds displayed positive to the 10 m riparian vegetation. As intact riparian vegetation within that buffer increased, the tolerant and omnivorous in the sampled reach decreased. As the proportion of intact riparian vegetation within the 0–10 m buffer increased, insectivorous and benthic insectivorous individuals also increased.
The 0–10 and 40–50 m riparian bands are representative of two scales of vegetative influence on stream fish communities. All results to this point emphasize the importance of near-stream vegetation. In a second, separate analysis (Table 6), the results for a 0–50 m percent riparian vegetation scale can be interpreted as the transitionary space between the two previously discussed vegetative buffers. All model averaged estimates across fish guilds displayed similar directional responses to riparian cover, while remaining unresponsive to the disturbed land-cover term. The strengths of the estimated relationships were less for the environmentally tolerant, omnivore, insectivore, and benthic insectivore guilds. The induced generalization from pooling riparian cover diluted otherwise strong relationships (compared to Table 5), further suggesting a difference in effect magnitude at different spatial scales.
Model results suggested the near-stream 0–10 m riparian vegetation is more influential on fish population structure than vegetation more distant from the stream. The mechanisms by which the vegetated buffer strip influences the fish community were not directly quantified, so our results are correlative and riparian conditions may be representative of multiple causal factors.
Riparian vegetative filtering of allochthonous sediments, nutrients, and other harmful adsorbed substances has been shown to have a positive influence on stream water quality, and prevents sedimentation of benthic fish habitat and stream eutrophication. Sediment and nutrient reduction are often related; as much as 70% of phosphorus and 20% of organic nitrogen can be bound to sediment particulate (Probst, 1985). The consequences of this are shown in Lee et al. (2003), where 7.0-m of switchgrass removed >92% of sediment, 80% total nitrogen, and 78% total phosphorus, and 16.5 m of a switchgrass – woody mosaic removed >97% of sediment, 94% total nitrogen, and 91% total phosphorus in Iowa streams. This suggests in-stream conditions may have been degraded in stream reaches with decreased levels of riparian cover.
Fish guilds have been shown to respond to differing nutrient and sediment levels. The environmentally tolerant guild was defined by the ability to persist in conditions associated with sedimentation or eutrophication (Niemela et al., 1998). The increased tolerant-species abundance in areas with decreased riparian cover suggests such conditions may have been present. The increased abundance of omnivorous species, or dietary generalists, in the more disturbed environments also suggests that riparian disturbance is linked to degraded fish habitat. Sedimentation removes the habitat and invertebrate diversity of agricultural streams (Schlosser and Karr, 1981). As conditions degrade and become more homogenous, regions previously occupied by specialist species become more prone to invasion by generalists (Marvier et al., 2004). The model-averaged estimates for both generalist guilds therefore suggest a link between decreased riparian vegetation and degraded fish habitat. Degraded in-stream habitat would give these species a competitive advantage and explain their increased abundance in regions with decreased riparian cover.
In areas with higher levels of intact vegetation within 10 m of the stream, insectivorous species percent composition was greatest, which suggested that in-stream habitat conditions were likely more heterogeneous. The insectivorous species are specialists, and habitat heterogeneity is usually related to specialist abundance (Marvier et al., 2004). The benthic insectivores, another specialist guild, are highly sensitive to sedimentation. Sedimentation destroys the complex benthic interstitial habitat on which their invertebrate prey relies (Niemela et al., 1998). The benthic insectivorous model-averaged estimate and 95% CI suggest a positive less-variable response to increasing levels of riparian cover, compared to the insectivorous guild. While a more consistent response to stream degradation is expected given the benthic insectivore sensitivity to sedimentation, the benthic insectivore guild was also less diverse (n = 15) than the insectivores (n = 36). The greater diversity of the insectivorous guild may have contributed to the more variable response. It is also worth noting that some of the most common insectivorous species were also environmentally tolerant, potentially indicating low water quality but sufficient habitat (Table 1). The estimates for both specialist guilds suggested a positive relationship between increased riparian vegetation and heterogeneous in-stream habitat.
Although no guild showed a response to the 40–50 m riparian vegetation or the HUC-12 landscape disturbance, guilds that evidenced a response to the 0–10 m intact riparian vegetation displayed similar responses to the full 0–50 m riparian vegetation. The estimates indicated a weakening fish-guild response to intact vegetation with distance from the stream, likely a result of ecological response scaling. Vegetation very near the stream (0–10 m) strongly influenced fish populations, and the influence became less as the scale of vegetation moved outward (0–50 m), eventually fading to zero (40–50 m and HUC-12). Because we did not evaluate scales of riparian vegetation intermediate to those discussed above (e.g. no models for 0–5 m or 0–25 m), we acknowledge that the proposed ecological scaling phenomena only addresses those buffer distances explicitly modelled.
Scaling of riparian mechanisms have been observed in an agricultural landscape where a 2-m buffer reduced total phosphorus by 31%, and an increase to a 15-m buffer reduced total phosphorus by 89% (Abu-Zreig, 2003). At the 2-m scale, each meter of vegetation reduced phosphorus 15.50%, while efficiency declined to 4.46%/m at 15-m. While major flow-related hydrological parameters, such as maximum depth and channel morphology, are related to catchment or basin level influences, sediment-related variables are better predicted by stream buffers (Richards et al., 1996). Considering the Red River of the North drainage basin's highly erodible clay and silt soils (Goldstein, 1995) and 71.5% row crop agricultural land use (Strong, 2010), the effects of riparian vegetation observed on the fish community may be driven by either allochthonous sediment or sediment-bound nutrient inputs.
The community response may also have been caused by a lack of allochthonous litter and arthropod inputs. While litter inputs vary by ecoregion and vegetative community, higher standing crops of herbaceous organic matter in streams are associated with diverse communities of shredder macroinvetebrates, including orders sensitive to degraded water quality (Plecoptera, Trichoptera) (Menninger and Palmer, 2007). Higher above-ground riparian vegetation biomass in a Wyoming trout stream was associated with higher terrestrial and aquatic arthropod biomass, as well as higher fish biomass. The increased riparian vegetation biomass was also associated with a higher biomass of arthropods in fish diets (Saunders and Fausch, 2007).
The increased abundance of insectivores in reaches with more riparian vegetation may then be a result of increased prey, just as insectivore absence might reflect a lack of prey. The trophic flow from riparian litter inputs to aquatic macroinvertebrates and eventually ending with predatory fishes is one possible mechanism underlying the associations observed here. Another may be the input of terrestrial arthropods at the riparian vegetation – stream interface also influences fish diets and, through a predation buffer for herbivorous aquatic arthropods, stream primary productivity (Nakano et al., 1999). A lack of riparian vegetation could reduce terrestrial arthropod inputs, and, in the study area, would result in a concurrent lack of filtering services, degrading the habitat and stream food web.
Although any of the proposed mechanisms could independently structure fish community composition, we believe it is more likely that a combination of effects is at work. In all cases, the ecological functioning is dependent on the 0–10 m of intact vegetative cover directly adjacent to the stream. While our understanding of the stream ecology may be incomplete, our evidence supports a link between intact 10-m buffer strips and a more diverse stream fish community.
Federal rules define 10.7-m of undisturbed vegetation on both banks as a fully-intact riparian buffer (Natural Resource Conservation Service, 2007), and Minnesota Statute 103 F.48 requires a 9.1-m minimum riparian buffer along streams. Substantial evidence suggests that water quality will respond positively in many cases to the treatments. Given the rapid rates of land conversion to production agriculture in the Midwest, care should be taken to protect specialist aquatic species and their habitat by ensuring adequate riparian buffering of streams. Our evidence indicated increases in the 10-m scale of intact riparian vegetation are strongly associated with specialist fish community composition, supporting this buffer as the minimum conservation goal in watershed restoration of low-gradient systems in landscapes dominated by agricultural land-use practices.
The authors thank Aaron Larsen, North Dakota Department of Health, and Scott Gangl, North Dakota Game and Fish Department for data sharing. Two anonymous reviewers provided valuable feedback and comments.
This study was funded by the University of North Dakota Department of Biology, Stella Fritzell and Joe K. Neal Memorials, and North Dakota View.